MTK Attribution Setup — Claude Skill

A Claude Skill that walks through installing MTK Attribution on a client site: pointing at the right platform install guide, confirming the field map, and adding the hidden form fields with exactly the right names.

What's inside

mtk-attribution-setup-claude-skill.md — the skill itself. It carries:

• Where the real docs live — the docs hub, all twelve platform install guides with their slugs and setup method, the reference pages, and troubleshooting. The skill is told to open the live guide before giving setup instructions, since platform steps change.
• The naming rules — the visible Label is human-readable and never prefixed with "MTK" (First Channel, not MTK First Channel), while the HTML name attribute must be exactly mtk_first_channel because the tag matches on it. Click ID and Meta cookie labels stay lowercase (gclid, fbc) — they're wire identifiers, not prose.
• A fetch step instead of a field list — the skill carries no copy of the fields. It fetches https://www.boggsmtk.com/products/attribution/docs/fields.md, raw Markdown rendered straight from the live field map, before naming anything. If the fetch fails it asks for the dashboard's Export CSV rather than guessing, and it treats a client's own field map as outranking the standard one.
• Add all of them by default — the whole standard set unless the user says otherwise, because a field that was never on the form has no history to answer with later and cannot be backfilled.
• Reusing fields that are already there — when a site already runs Attributer or already has hidden utm_* inputs, work out what each existing field corresponds to, then rename it to MTK naming rather than freezing the old name in place. Carries the Attributer field/token → MTK mapping, the loose-UTM mapping, how to apply a rename per platform (including the Default-Value platforms where the name isn't yours to set), the CRM re-pointing that a rename implies, and the value-shape changes that survive it (channel wording, landing page as a path). The Attributer drilldowns are treated as a starting guess to confirm with the user, not a mapping — what lands in them depends on which UTMs that client actually sends.
• A final ordering pass — First, then Last, then IDs, then Insights, on the form and in the field map, so a lead record reads as one first-touch story then one last-touch story instead of alternating dimension by dimension.
• Verification steps and the most common failure (a name mismatch producing a silently empty field).

mtk_attribution_summary (the all-fields roll-up) and mtk_journey_json (the machine-readable journey) are listed in the fetched reference under "Not in the standard set" — real fields, just not ones a form gets by default.

Installing it in Claude

Claude expects a skill's entry file to be named SKILL.md inside a folder named for the skill. To install:

mtk-attribution-setup/
└── SKILL.md      ← mtk-attribution-setup-claude-skill.md, renamed

The YAML frontmatter at the top of the file (name, description) is what Claude reads to decide when to trigger the skill — keep it intact.

Keeping it current

This skill is meant to be handed to customers, who run it in their own Claude. A wrong detail produces an empty field on a client's form with no error anywhere, in an account nobody here is watching — so field names are fetched, not written down. Do not paste the field list back into the skill. The moment a field is added or renamed, every installed copy would start handing out names that no longer match.

Adding, renaming, or repointing a field needs no change here at all: the reference is rendered from DEFAULT_FIELD_MAP and FIELD_TITLES in src/lib/mtk.ts (whboggs/boggsmtk) on every deploy. Same for a new engine data source (whboggs/mtk-attribution, src/engine.js).

What still needs updating by hand:

• a docs page added, removed, or moved — the URLs and platform slugs are hardcoded in the skill, /products/attribution/docs/fields.md included
• a platform changing its install method
• Attributer renaming a field or changing a token, which breaks the remapping table (https://help.attributer.io)

The skill file repeats this in its own "Keeping this skill current" section, so whoever opens it sees the obligation too.